RM新时代有限公司

Re-ID commercial level

CloudWalk Technology's cross-domain tracking (Re-ID) technology has broken three world authoritative records.
With accuracy reaching commercial level, artificial intelligence has entered a new era from "face recognition" to "person recognition".

Librispeech World Record

CloudWalk Technology's speech recognition refreshes the Librispeech world record.

OCR refreshes international leaderboards.

CloudWalk Technology's natural scene OCR on the international authoritative dataset ICDAR
has refreshed multiple global best results.

Re-ID World Record

CloudWalk Technology sets three new world records in cross-domain tracking (Re-ID).

Major Breakthrough in Natural Language Processing

CloudWalk Technology, in collaboration with Shanghai Jiao Tong University, achieved a major breakthrough in natural language processing and machine reading comprehension.

Surpassing the human score of 69.4% on high school test questions.

Breakthrough in 3D Human Reconstruction

CloudWalk Technology achieves breakthrough in 3D human reconstruction technology.
Accuracy error reduced by 30%, algorithm speed increased by dozens of times, reaching 5 milliseconds.

Papers on machine reading comprehension and image recognition accepted by AAAI.

CloudWalk Technology's papers on machine reading comprehensionand image recognitionhave been accepted by AAAI, a top international academic conference in the field of artificial intelligence.AAAI 2020accepted.

Re-ID Cross-Domain Tracking Technology Sets World Record Three Times

CloudWalk Technology's Re-ID technology achieved the best results on three authoritative mainstream evaluation datasets in the industry: Market-1501, DukeMTMC-reID, and CUHK03-NP. This is another refresh of the Re-ID world record by CloudWalk after 2018 and 2019.

Won Double Championships in Global Authoritative FRVT Test

In the world's most authoritative face recognition algorithm test (FRVT), CloudWalk Technology's team won two championships and one runner-up in three tracks: face recognition 1:1, face recognition 1:N, and face recognition with mask occlusion.

Speech Technology Refreshes Multiple Authoritative Records

CloudWalk Technology's speech recognition, semantic error correction, deep learning noise reduction, and other technologies have refreshed multiple international and domestic authoritative speech recognition records, including the authoritative Chinese speech recognition dataset Aishell, Tsinghua University's Thchs30 test set, and the international top conference Interspeech 2020 DNS Challenge dataset.

Passed Trusted Cloud Evaluation Service

In August 2021, the China Academy of Information and Communications Technology announced the results of the first half of the year's Trusted Cloud evaluation. CloudWalk Technology's AI open platform, with its high-quality and reliable cloud service technology, passed the Trusted Cloud evaluation for three intelligent cloud application services: face recognition, speech recognition, and text recognition.

Won Championship

In December 2022, in the latest list of the global authoritative face recognition algorithm test (NIST-FRVT), CloudWalk Technology won all championships in three tracks: face recognition 1:1, face recognition 1:N, and face recognition with mask occlusion.

Refreshed World Record

In July 2023, CloudWalk Technology's pedestrian foundation large model refreshed world records on four datasets: PA-100K, RAP V2, PETA, and HICO-DET.

Champion of the ICCV 2023 Fine-Grained Activity Detection Challenge

CongRong won the OpenFAD23-ICCV23 behavior classification track, scoring 93.87% on coarse-grained and 91.96% on fine-grained activity recognition.

New Result on the COCO Benchmark

The CongRong visual foundation model achieved a new result on COCO-related evaluation, demonstrating generalization and low-data transfer capabilities.

Third Globally and First in China on OpenCompass

In the June 2024 OpenCompass multimodal evaluation, CongRong averaged 65.5, ranking third globally and first in China, with an OCRBench score of 827.

CoSEFA Accepted by FSE 2025

CoSEFA, a programming agent jointly developed by CloudWalk and Chongqing University, was accepted by the ACM SIGSOFT Symposium on the Foundations of Software Engineering 2025.

CongRong Ranked First on the May 2025 OpenCompass Multimodal Leaderboard

In May 2025, CongRong scored 80.7 and ranked first on the then-current OpenCompass global multimodal leaderboard. The result is time- and benchmark-specific.

UAV-Anti-UAV Multimodal Tracking Benchmark Released

A joint research team released the million-scale air-to-air multimodal UAV tracking benchmark. MambaSTS achieved the best performance across all five benchmark metrics.